Analysis question about priors

[colon-perez]

Hello,

I'd like to know more about the priors in the analysis portion, which is a mandated selection that defaults to mouse results; however, I am working with rats and would like to get some insights to make better decisions. The preprocessing and confound are running well, and the analysis too, but I noticed by reading the help about these options that I was leaving on the default setting
--prior_maps PRIOR_MAPS (default: /home/mambauser/.local/share/rabies/melodic_IC.nii.gz)
--prior_bold_idx [PRIOR_BOLD_IDX ...] (default: [5, 12, 19])
--prior_confound_idx [PRIOR_CONFOUND_IDX ...] (default: [0, 2, 6, 7, 8, 9, 10, 11, 13, 14, 21, 22, 24, 26, 28, 29])

Do you think these selections are good enough for rat analysis? How were these selected? I understand these are outputs from Melodic, but what are those specific components, and what made those three bold and the remaining confounds? I am just observing the results, and they don't look bad, but I now wonder if they are correctly done.

[Gab-D-G]

Hi,

You are right that these are ICA components obtained from mouse data, and are not compatible with rats. If you do not run dual regression, or the --data_diagnosis, these are not used, and so those parameters will be inconsequential. Otherwise, if you want to use those options, you need to provide the path (with --prior_maps) to a ICA decomposition from rat data that overlaps with your dataset in commonspace. You could do a group ICA decomposition within RABIES to get that from your own data.

If you are curious about those parameters, consult the details in the QC webpage https://rabies.readthedocs.io/en/latest/analysis_QC.html, and the default ICA components are displayed as a supplementary figure in the rabies paper https://www.nature.com/articles/s41467-024-50826-8.

[Gab-D-G]

You are correct that ReHo and ALFF are not implemented. The data in confound_correction_datasink/cleaned_timeseries is indeed the fMRI data post-denoising, which you can use for whichever downstream application you like. You can use the brain mask from the bold_datasink folder as well if you need a brain mask to extract the timeseries from the cleaned_timeseries nifti files.

[colon-perez]

I appreciate all the answers! I have one last question (at least I hope for now). For the ReHo, it seems straightforward to use 3dReHo from AFNI and also 3dRSFC for ALFF, but I want to make sure that the preprocessing does not apply the censoring directly to the cleaned time series (see here: https://neurostars.org/t/alff-falff-calculate-manually-with-python3/30516; it actually is a common caveat from AFNI to avoid censoring data and apply bandpass filters, which I am not). I found a CSV file in the confound stating the censored volumes, so I presume this is used for the analysis, but not applied directly to the cleaned timeseries. Do you think this is a correct interpretation?

[Gab-D-G]

If you have selected the frame censoring option during the confound_correction stage, then the output data will be missing frames, which does make it incompatible with any frequency estimator that cannot account for missing frames. Usually that's the case for ALFF. So you probably should not apply censoring if you're going for that analysis approach.
Note that within RABIES we deal with this issue by simulating missing data points while preserving the frequency profile of the timeseries here

RABIES/rabies/confound_correction_pkg/confound_correction.py

Line 333 in 1764266

timeseries_filled = lombscargle_fill(x=timeseries,time_step=TR,time_mask=frame_mask)

. If you want, you could hack something together in python using that same function, and the censoring CSV, to fill the missing points prior to ALFF estimation. I am not sure it was done before for ALFF specifically, but this is something that is done for combining censoring and bandpass filters.

[colon-perez]

Sounds good! I am not using censoring in confound correction so it should be ok. I will look at that function as well. I appreciate all the input.

[Gab-D-G]

I'm happy to help!